Effects of functional traits on the prediction accuracy of species richness models

Abstract

AbstractAimWe assess main ecological determinants affecting the comparative performance of macroecological models (MEMs) that model species richness directly, and stacked species distribution models based on stacking probabilities (pS‐SDMs) and binary predictions (bS‐SDMs). Specifically, we aimed to understand how statistical effects such as prevalence and environmental heterogeneity are entangled with species' ecology in Swiss avian assemblages.LocationSwitzerland.MethodsWe tested for statistical and ecological effects on overprediction and underprediction by regressing species richness residuals against community‐averaged values of prevalence, functional traits and functional dissimilarity. Further, we defined bird functional groups through hierarchical clustering and compared accuracy of species richness predictions between groups to understand the differences between model types and ecological determinants thereof. Last, we tested how accuracy of species assemblages constructed from bS‐SDMs relates to species' functional characteristics.ResultsUnderprediction of high diversity sites by pS‐SDMs and MEMs was mainly explained by prevalence, whereas overprediction of low diversity sites was strongly affected by diet and habitat traits, and increased with functional dissimilarity. Model performances varied strongly between functional groups with more accurate and less biased predictions for generalist species groups. Critically, overprediction bias in richness predictions by bS‐SDMs was uncorrelated with assemblage prediction success.Main conclusionsThe reliability of all community models tested here strongly depended on functional species' characteristics related mainly to diet, foraging and breeding habitat. This underlines the need to incorporate all relevant and species‐specific or functional group‐specific ecological filters in the models. Improved prediction accuracy of species richness will require finer‐resolved environmental predictors that better describe available niche space and account for specific spatial and resource requirements of different species. More research is needed to understand the relationship between accuracy of species richness and species assemblage predictions in bS‐SDMs as well as the role of biotic interactions.